1 |
Thank you very much for sharing. |
2 |
|
3 |
P.V.Anthony |
4 |
|
5 |
On this day, 03-August-2007 1:41 AM, Steve Herber wrote: |
6 |
> The point of Unix has always been to manage resources. In the old days, |
7 |
> where you only had a single CPU, the kernel had to time share among all |
8 |
> the different processes. The ps command shows you all the different |
9 |
> processes. Your system will have many processes. |
10 |
> |
11 |
> Once you go to a multiple-cpu system you actually get to take advantage |
12 |
> of the original Unix choice of fork/exec to create new processes. Each |
13 |
> process can run on a different cpu core. Now instead of timesharing a |
14 |
> single core across every process, you can have 2 or 4 processes running |
15 |
> on their own core. |
16 |
> |
17 |
> In you sample below, I could see apache running on one core, mysql on |
18 |
> another, postgres on another, while the applications written in perl, |
19 |
> ruby, and php running on the 4th. Of course you will really have |
20 |
> hundreds of processes running at a time so all 4 cores will have to |
21 |
> timeshare. As others said, you don't get 100% from each new core. |
22 |
> There is overhead from the kernel to figure out where to put a process |
23 |
> for the next slice and there will be more memory contention because you |
24 |
> have 4 CPU's talking to memory. A 4 core system will have almost twice |
25 |
> the cpu power of a 2 core system. Most of your processes look like they |
26 |
> will be IO bound but when a CPU bound tasks runs, it will only take over |
27 |
> one core leaving the other three to handle the IO bound ones. |
28 |
> |
29 |
> Your initial issue is cost and rightly so. A 2 core system is cheaper |
30 |
> than a 4 core system. If you build equivalent 2 core and 4 core systems |
31 |
> then you could test them against you real workload and see if the 2 core |
32 |
> system is better. You might discover that you should really have 2 two |
33 |
> core systems for the load you have and for redundancy. Or you might |
34 |
> discover that you really need to address the IO bottleneck and get |
35 |
> multiple SCSI data paths. I would buy a 4 core system with lots of |
36 |
> disks. Then I would run a test to benchmark my system, probably just |
37 |
> running top to see the 15 minute load average. Then I would use the cpu |
38 |
> control software to disable one, two, and then three cores while my |
39 |
> normal load was running and I would again look at the 15 minute load |
40 |
> average. I would expect to see it rise as the number of CPU's was |
41 |
> reduced. My next computer would have the number of cores that met the |
42 |
> sweet spot. |
43 |
> |
44 |
> Cheers! |
45 |
> |
46 |
> Steve Herber herber@×××××.com work: 206-221-7262 |
47 |
> Security Engineer, UW Medicine, IT Services home: 425-454-2399 |
48 |
> |
49 |
> On Thu, 2 Aug 2007, P.V.Anthony wrote: |
50 |
> |
51 |
>> On this day, 02-August-2007 4:58 PM, Pascal BERTIN wrote: |
52 |
>>> P.V.Anthony a écrit : |
53 |
>>> |
54 |
>>> <snip/> |
55 |
>>> > Apps. |
56 |
>>> > 1. Gentoo linux 64bit |
57 |
>>> > 2. Apache 2 |
58 |
>>> > 3. MySql |
59 |
>>> > 4. Postgres |
60 |
>>> > 5. Qmail |
61 |
>>> > 6. Pure-ftpd |
62 |
>>> > 7. Mod_perl |
63 |
>>> > 8. php |
64 |
>>> > 9. ruby |
65 |
>>> > |
66 |
>>> <snip/> |
67 |
>>> |
68 |
>>> aren't you forgetting the app that you will run the most : |
69 |
>>> emerge. |
70 |
>>> |
71 |
>>> This one will really benefit from quad core, and during time where you |
72 |
>>> will work on the server. |
73 |
>>> So, facing such a choice, I would also consider my comfort/time, and |
74 |
>>> give quad core a +1. |
75 |
>>> |
76 |
>>> Pascal |
77 |
>> |
78 |
>> What I am really interested is, how the server will perform most of |
79 |
>> the time. Will the instance of the apps go to each core in a balanced |
80 |
>> way? |
81 |
>> |
82 |
>> From the previous posts, I gather that the amd chip is really good. |
83 |
>> |
84 |
>> I must say that I love the intel drivers for the network and sata. If |
85 |
>> only amd has some chip sets for their own cpus and good open source |
86 |
>> drivers for their chip sets. Intel has that covered and I think the |
87 |
>> intel drivers are open source. |
88 |
>> |
89 |
>> Please correct me if I am wrong. |
90 |
>> |
91 |
>> P.V.Anthony |
92 |
>> |
93 |
>> -- |
94 |
>> gentoo-amd64@g.o mailing list |
95 |
>> |
96 |
> |
97 |
> !DSPAM:6,46b21781197995167217508! |
98 |
|
99 |
-- |
100 |
gentoo-amd64@g.o mailing list |